Michael Burry has walked back into the AI trade, and once again he is entering through the emergency exit.
The investor made famous beyond Wall Street by The Big Short has been building and reinforcing bearish positions around some of the biggest beneficiaries of the artificial intelligence infrastructure boom. His targets have included Nebius, Micron, Oracle and the iShares Semiconductor ETF (SOXX). At the same time, Burry has been adding to bullish positions elsewhere, including MercadoLibre and animal-health company Zoetis.
That combination matters. This is not simply Michael Burry declaring that AI is fake, useless or destined to disappear. It looks much more like a valuation trade: some assets have become too expensive relative to the cash flows and returns Burry thinks they can ultimately produce, while neglected companies elsewhere look comparatively attractive.
Nebius has become an especially provocative target. Burry has described the AI infrastructure company as the kind of thing you see at the “top of a boom.”
And that gives investors a better question than the usual one: Is Michael Burry right again?
The more useful question is this: how do you tell the difference between a legitimate industrial boom and a financial bubble built on legitimate technology?
Table of contents
- The Big Short Guy Has Entered the Chat
- Why Nebius Triggered His Bubble Radar
- But This Isn’t an “AI Is Fake” Trade
- Cassandra Has a Timing Problem
- So What? Don’t Copy the Character, Read the Signal
- Michael Burry and the AI boom: key questions
The Big Short Guy Has Entered the Chat
Burry’s cultural problem is that he is no longer just an investor. He is a character.
For millions of people, Michael Burry is still Christian Bale sitting in an office, blasting music and waiting for a housing thesis nobody else believed to finally become obvious. That history means every new bearish bet arrives with cinematic baggage attached.
His latest AI shorts deserve a less dramatic reading.
Burry has targeted several different layers of the AI capital-spending machine. Micron sits inside the memory-chip boom. SOXX provides broad exposure to the semiconductor industry. Oracle has committed enormous resources to cloud infrastructure. Nebius is essentially a concentrated bet on demand for AI compute.
The scale of the infrastructure race is difficult to ignore. Oracle, for example, reported that fiscal 2026 cloud infrastructure revenue rose 77% to $18.1 billion. But the company also generated negative $23.7 billion in free cash flow during the fiscal year as it invested aggressively in cloud capacity. Oracle said it raised $43 billion in debt financing and $5 billion in equity financing during fiscal 2026.
Those numbers can support two completely different narratives.
The bullish version says AI demand is so strong that companies must invest enormous amounts today to capture a once-in-a-generation computing transition.
The bearish version says precisely the same thing — except it asks whether the eventual returns will justify the money being spent.
That distinction is where the interesting part begins.
Why Nebius Triggered His Bubble Radar
Nebius almost looks engineered to test an investor’s feelings about the AI boom.
The Amsterdam-based company is building AI-focused cloud infrastructure around large GPU clusters, data centers and related software. And the underlying business is not imaginary. Nebius has been posting explosive growth while rapidly expanding capacity.
In the first quarter of 2026, the company reported approximately $2.5 billion in capital expenditures, driven primarily by GPUs, GPU-related hardware and data-center expansion. Its Nebius AI cloud operation generated adjusted EBITDA of $174 million with a 45% adjusted EBITDA margin during the quarter.
In other words: this is not a dot-com company with a URL, a Super Bowl ad and no business model.
But real growth does not make valuation irrelevant.
AI infrastructure is brutally capital intensive. GPUs have to be purchased. Data centers have to be built or leased. Electricity has to be secured. Networking hardware must be installed. Financing has a cost. Meanwhile, computing hardware improves at a pace that creates another awkward question: how long will today’s extremely expensive infrastructure remain economically competitive?
That is one of the fault lines in the AI bubble debate.
When investors value an AI infrastructure company, they are not merely predicting demand. They are implicitly making assumptions about utilization rates, pricing, depreciation, financing costs, customer concentration, future competition and the useful economic life of hardware.
A company can therefore experience spectacular revenue growth and still become a terrible investment if the stock price assumes something even more spectacular.
This is what “top of a boom” language is really getting at. A boom becomes dangerous not when the underlying technology stops working, but when expectations become so aggressive that excellent execution is no longer enough.
But This Isn’t an “AI Is Fake” Trade
Reducing Burry’s position to “Michael Burry thinks AI will fail” would miss the point.
His portfolio activity has also included buying companies he believes the market has treated less generously. Earlier in 2026, Burry added to MercadoLibre and Zoetis, among other positions. He described MercadoLibre as a long-term opportunity after a selloff and increased his exposure to Zoetis as well.
That makes the broader message closer to price matters than technology does not.
This distinction is essential for investors because financial bubbles routinely form around technologies that actually change the world.
The internet was real in 1999.
E-commerce was real.
Fiber-optic networks were useful.
Personal computing was transformative.
None of that prevented investors from paying absurd prices for some businesses attached to those trends.
AI could follow a similar pattern without following the same script. Artificial intelligence may transform software, search, advertising, medicine, industrial automation and knowledge work while some AI-linked stocks still deliver disappointing investment returns.
Technological success and shareholder returns are related, but they are not identical.
The stock market does not ask whether AI will matter. At current valuations, it increasingly asks how much AI will matter, how quickly revenues will arrive, who will capture the profits and how much capital must be burned to get there.
That is a much harder model.
Cassandra Has a Timing Problem
Burry’s own Substack is called Cassandra Unchained, which is almost too perfect.
In Greek mythology, Cassandra could foresee disaster but was cursed not to be believed. For short sellers, there is an additional curse: seeing the danger is not enough. You also need something approaching the right timing.
An overvalued stock can become more overvalued.
Then dramatically more overvalued.
Then completely ridiculous.
And only then collapse.
That sequence is lethal to anyone who assumes that identifying excess automatically produces a profitable short.
Short selling creates asymmetry that buying a stock does not. A long investor can theoretically lose the capital invested if a stock goes to zero. A short seller faces potentially unlimited losses because a stock has no theoretical ceiling. Borrowing costs, margin requirements and volatility can force a trader out before the thesis ever plays out.
Burry understands this better than almost anyone. His housing trade became legendary precisely because being early was extraordinarily painful.
The lesson from The Big Short was never simply “spot the bubble.” It was also that markets can remain detached from your estimate of fundamental value for an uncomfortable amount of time.
That matters in AI because the bull case still has serious evidence behind it. Oracle’s fiscal 2026 cloud infrastructure revenue nearly doubled, while its remaining performance obligations reached $638 billion. Nebius is scaling rapidly. Demand for accelerators, memory, power and data-center capacity has produced genuine revenue across the supply chain.
A bubble thesis therefore has to survive the possibility that the boom gets fundamentally stronger before financial excess becomes visible.
Cassandra may see Troy burning. The portfolio still needs to survive until somebody lights the match.
So What? Don’t Copy the Character, Read the Signal
Investors do not need to copy Burry’s shorts to take his warning seriously.
In fact, blindly following a famous contrarian is arguably the least useful lesson available here. You do not know his complete portfolio, hedges, position sizing, time horizon, entry prices or willingness to absorb losses.
What you can copy is the skepticism.
Before buying an AI-linked stock because the ticker has become associated with the hottest technology cycle on Earth, ask a few less exciting questions.
How much growth is already priced in? A company growing revenue 50% annually can still disappoint if its valuation assumes 80%.
What happens to margins as competition increases? Scarce GPU capacity can command premium economics. Those economics may change when supply expands.
How much capital is required to generate each additional dollar of revenue? Revenue growth funded by enormous capital expenditure deserves different treatment from highly scalable growth funded largely by existing infrastructure.
How long will the hardware remain economically productive? If new generations of accelerators dramatically improve performance per dollar or per watt, depreciation assumptions become central to the investment case.
Who are the customers? Huge contracts look reassuring, but customer concentration can transform one delayed project or lost account into a major problem.
And finally: what has to go right to justify today’s price?
That may be the most important question in the entire AI trade.
There is a massive difference between a company that can succeed under several scenarios and one whose valuation requires near-perfect execution.
Michael Burry could be early again. He could be wrong. The AI infrastructure boom could run much further than its skeptics expect. Or the industry could eventually discover that it built too much capacity, borrowed too aggressively and priced too much future growth into the present.
None of those outcomes requires AI itself to fail.
That is the uncomfortable signal behind Burry’s latest trade. The biggest bubbles are often built around something real. The dangerous moment comes when investors stop asking what is this business worth? and start asking only how much higher can it go?
The character from The Big Short may make the headline. The valuation problem is the story worth watching.
Sources: Michael Burry’s Cassandra Unchained, Oracle FY2026 results and Nebius.
Michael Burry and the AI boom: key questions
What AI stocks is Michael Burry shorting?
Burry has disclosed bearish positions across the AI infrastructure trade, including Nebius, Micron, Oracle and the iShares Semiconductor ETF (SOXX), alongside other AI-related bearish bets during 2026. Individual positions can change quickly.
Does Michael Burry think artificial intelligence is a bubble?
Burry has repeatedly warned about speculative excess surrounding AI and compared aspects of the current market with previous bubbles. That does not necessarily mean he believes AI technology itself has no value.
Why is AI infrastructure spending potentially risky?
Building AI infrastructure requires enormous spending on GPUs, data centers, networking and electricity. Investors must determine whether future revenue and margins will produce adequate returns on that capital.
Can an investor be right about a bubble and still lose money?
Yes. A stock can remain overvalued or become substantially more expensive before falling. Timing is particularly important for short sellers because losses can grow as prices rise.
What should investors check before buying an AI stock?
Look at valuation, revenue growth, margins, free cash flow, capital expenditure, customer concentration, debt and the assumptions about future growth already embedded in the share price.
Is this article financial advice?
No. This article is for informational purposes only and does not constitute investment, trading or financial advice. Investors should conduct their own research and consider their financial circumstances and risk tolerance.